Local Qualitative Shape from Stereo without Detailed Correspondence Extended Abstract

Schemes for the extraction of qualitative shape information from stereo tend to relegate the solution of the correspondence problem to a preprocessing stage and to assume that their input is provided in the form of a matched image pair { a disparity map [1]. The disparity map is a rich source of shape information, which allows the recovery of depth on a ratio scale [2] even when the camera geometry is unknown. Consequently, using the disparity map to extract qualitative rather than inexact quantitative shape information may be well-justi ed by considerations of stability and robustness [3], but appears to be wasteful in the sense that a rich and di cult to compute representation is transformed into a relatively terse one (Figure 1).